868 results about "Completion time" patented technology

The invention discloses an intelligent engineering vehicle scheduling method based on artificial intelligence, and the method comprises the steps: constructing a scheduling objective function which at least comprises the engineering vehicle transportation cost, the task completion time cost, the path congestion cost and the energy consumption cost; establishing a multi-dimensional task demand prediction model based on historical task data and real-time traffic data, and outputting a task distribution prediction value and a traffic state prediction value in a future time period; inputting the model into an engineering vehicle dynamic scheduling model established based on a deep reinforcement learning algorithm for iterative optimization; and generating a real-time scheduling instruction based on the optimized dynamic scheduling model, dynamically allocating task paths and resources of the engineering vehicle, and monitoring an execution state in real time to adjust a scheduling strategy. According to the method, efficient, economical and environment-friendly intelligent scheduling of the engineering vehicle is realized through construction of a scheduling objective function, multi-dimensional demand prediction, deep reinforcement learning dynamic optimization and real-time scheduling and monitoring.

The invention provides a discrete MES-oriented intelligent production scheduling system, method and equipment and a medium, and belongs to the technical field of discrete manufacturing industry production scheduling. Data is acquired through a sensor and serves as production scheduling data; establishing a material inventory data association order ID and establishing an index; determining a process sequence constraint, a calculation equipment productivity constraint, a material supply constraint and an order priority constraint; initializing a population based on a genetic algorithm, randomly generating N groups of process sorting schemes, calculating a utilization rate index, and taking a comprehensive score as a fitness value; outputting a better solution set; the optimal solution of the genetic algorithm is used as initial pheromone distribution, high-quality path pheromones are enhanced according to the actual production effect of the completion scheme, and if the preset number of iterations is reached, the operation is stopped, and an optimized production scheduling scheme is output; and checking the production scheduling plan through a graphical interface. Through continuous optimization of the procedure sorting scheme, the equipment utilization rate is effectively improved, the total order completion time is shortened, the production resource configuration is optimized, and the production efficiency is improved.

The invention provides a resource scheduling control method and system for a big data server, and the method comprises the steps: constructing a multi-dimensional resource portrait module, collecting the CPU, memory, network, storage I / O load and task queue length of each node in real time, and predicting a resource demand trend through a time sequence algorithm; extracting characteristics such as calculation intensity, data dependence, memory requirements, network transmission quantity and the like; adjusting the weight coefficients of the resource utilization rate, the task completion time and the energy consumption efficiency according to the system load and the historical effect; establishing a bipartite graph model by taking a resource trend as a node feature and a task vector as an edge feature, and calculating a matching score through graph convolution and a multi-objective optimization function; the scheduling scheme is synchronized by adopting a consistency algorithm; automatic rollback and reallocation are carried out when resources are detected to be insufficient; and optimizing a weight coefficient and a network parameter through reinforcement learning. Through the method, the system resource utilization rate can be improved, the task execution efficiency is improved, the overall scheduling effect stability is improved, and the system fault recovery time is shortened.

The invention discloses an AI-based intelligent factory dynamic production scheduling optimization method and system, and belongs to the technical field of factory dynamic production scheduling, and the method comprises the following steps: obtaining and integrating the working states of various types of equipment in a factory, the stock position states of different materials, and the manual data; the production order state, the process route requirement, the plan and the completion time of each process and the delivery date requirement of the customer are determined; generating an initial production scheduling plan through an AI optimization algorithm in combination with supply chain material data, factory storage space, equipment switching cost, production rules, each order process dependency relationship and a production target; according to the method, the initial scheduling plan is generated by acquiring and integrating multiple types of production data, the dynamic events are monitored, multiple schemes are generated through evaluation, the optimal scheme is selected through economic model evaluation, and production refinement, dynamic response and benefit optimization are achieved.

The invention relates to an intelligent manufacturing management method and equipment based on a digital twinning technology, and a medium. The method comprises the following steps: constructing a collaborative model, collecting real-time data based on the collaborative model, establishing a manufacturing resource intelligent configuration simulation model, and generating an initial scheduling scheme; the initial scheme is optimized through a genetic algorithm, an executable scheduling scheme is generated, and the optimization target is to minimize the maximum completion time; applying the executable scheme to a workshop, monitoring a dynamic event and collecting data; when a dynamic event occurs, generating a rescheduling scheme based on an event-driven rescheduling strategy and Monte Carlo tree search; the rescheduling scheme is transmitted to a workshop, tasks are adjusted, and meanwhile process execution data are recorded; constructing a digital twinborn model based on recorded data, analyzing dynamic event influence, identifying key nodes in combination with a fault propagation network, and optimizing a dynamic scheduling strategy; and an optimized scheduling parameter and a model updating strategy are generated by using a digital twin simulation result, and closed-loop management and real-time response are realized.

The invention relates to an unmanned system cluster distributed task cooperative scheduling method and system. The system comprises a ground station general control scheduling module, an unmanned aerial vehicle cluster and a distributed monitoring and scheduling agent module. The method comprises the following steps: calculating the static priority of each task in a DAG task graph; tasks with high priorities are scheduled firstly; sequentially selecting a node with the minimum power consumption perception completion time EA-EFT from all the unmanned aerial vehicles, and allocating a task to the node; each monitoring scheduling node monitors the task execution progress in real time; and after the node fails, task allocation is stopped, the tasks are rescheduled according to the priorities of the tasks, the optimal allocation nodes of the tasks are recalculated based on an EA-EFT model, and a recommendation mechanism is triggered to take over the tasks of the failed nodes after each scheduling fails. The problems that an unmanned aerial vehicle cluster scheduling scheme is poor in adaptability, low in efficiency and incomplete in fault-tolerant mechanism are solved, and the scheduling efficiency, the system robustness and the energy efficiency of an unmanned aerial vehicle cluster in a complex task scene are improved.

The invention belongs to the field of workshop dynamic scheduling, and provides a dynamic disturbance-oriented tire vulcanization workshop intelligent flexible scheduling method and system, and the method comprises the steps: building a production scheduling model with the minimization of the maximum completion time as an optimization target; obtaining an initial processing sequence and a selectable machine set of each workpiece process, and solving by taking the production scheduling model as a target function to obtain an optimal initial scheduling scheme; when a dynamic event occurs in the production process, detecting the type of the dynamic event and the occurrence time of the event, and extracting a completed process, a processing process and an unstarted process; and determining a to-be-scheduled process based on the influence of the dynamic event on the processing process and the unstarted process, retaining the process processing sequence of the optimal initial scheduling scheme, updating the optimal processing machine of the to-be-scheduled process, and generating a rescheduling scheme. The device only replaces a machine in an unprocessed process, is embedded into a cooling time buffer area between the same workpiece processes, solves the problem of insufficient cooling of the rubber material, and guarantees the quality and process stability.

The invention discloses a multi-agent distributed task collaboration method and system, and relates to the technical field of computers, and the method comprises the steps: broadcasting a to-be-allocated task set to each agent, enabling the agents to estimate the task expected completion duration of each to-be-allocated task, and generating a corresponding to-be-allocated task expected duration matrix; receiving a to-be-allocated task expected duration matrix and a historical allocation task estimated duration from each agent; constructing a constraint optimization model according to the to-be-allocated task expected duration matrix corresponding to each agent and the historical allocation task estimated duration; and solving the constraint optimization model through a global constraint optimization algorithm, and allocating each task to be allocated to a corresponding agent according to an optimal solution. Therefore, the flexibility of task allocation is improved, and resource waste and unreasonable allocation in task allocation are reduced.

The invention relates to a clustering and entropy-guided reentrant hybrid flow shop scheduling method. The method comprises the following steps: step 1, establishing a problem model; step 2, setting algorithm operation parameters; 3, adopting an initialization strategy to generate an exploration population and a development population; 4, judging whether a first-stage termination condition is met or not, if not, executing a first-stage evolutionary strategy and an updating strategy on the exploration population and the development population, and otherwise, executing the step 5; 5, constructing an elite population; 6, judging whether a second-stage termination condition is met or not, and if not, executing a second-stage evolutionary strategy on the elite population; otherwise, outputting a Pareto solution set; and 7, updating the elite population. According to the method, dynamic balance of global exploration and local development is realized, and the solution distribution can be improved while the solution set convergence is ensured, so that the completion time and the total energy consumption are reduced, the production cost is reduced, and the workshop scheduling efficiency is improved.

The embodiment of the invention discloses a flexible job shop scheduling method based on preference-driven graph reinforcement learning, and relates to the field of shop dynamic scheduling in an intelligent manufacturing technology. According to the method, by constructing a multi-objective optimization model, multiple objectives of the job shop can be optimized at the same time, including the minimum completion time, the total delay and the total cost. Wherein an imperfect maintenance model is constructed, and the maintenance demand and the maintenance opportunity of each machine are dynamically determined. And capturing a complex relationship between the operation and the machine by using an improved graph neural network. In combination with a preference-driven mechanism, a maintenance plan and workshop scheduling are adjusted in real time through a graph reinforcement learning method, and efficient priority scheduling rules under different preferences are learned, so that an integrated decision of machine allocation, an operation sequence and maintenance arrangement in dynamic scheduling is realized.

The invention relates to the technical field of security auditing, in particular to a security auditing method and system for an industrial control system, and the method comprises the following steps: aiming at a key control task, a communication task and a security task of a real-time operating system, collecting a period, a starting timestamp, a finishing timestamp, a central processing unit occupied time slice and peak memory usage amount data. According to the method, the period, timestamp, central processing unit occupation and memory usage data of a key task of a real-time operating system are collected, a task execution time boundary and a resource consumption envelope are set, and then an expected relation rule set of a task time sequence and resource consumption is constructed by applying historical data statistics and logic rule deduction; and meanwhile, the key length of symmetric and asymmetric encryption, initialization vector generation, hash algorithm selection, key derivation parameters and encryption operation context are stipulated, so that a comprehensive and specific ICS behavior specification baseline is established.

The invention discloses a port tallying operation management method and system, and the method comprises the steps: dividing port tallying operation scenes, and defining and quantifying the node features of each operation node in each operation scene; constructing a dynamic monitoring strategy model, inputting the operation node features into the dynamic monitoring strategy model, and obtaining a monitoring strategy matched with the operation nodes; constructing a port tallying operation directed graph, and training and generating a node completion time window prediction model based on the constructed port tallying operation directed graph; obtaining operation data of each current operation node according to a dynamic monitoring strategy, inputting a node completion time window prediction model, and obtaining prediction completion time of each operation node; and for each operation node, predicting completion time, comparing preset time corresponding to each operation node, and performing operation node progress early warning according to a comparison result. According to the application, precise intelligent monitoring and early warning of port tallying operation can be realized, and the management quality is effectively improved.

The invention discloses a hybrid heterogeneous cloud workflow scheduling method based on reinforcement learning, and belongs to the technical field of cloud computing. The method comprises the following steps: aiming at a cloud workflow scheduling problem, by taking minimization of completion time as a target and taking cost and resources as constraints, a three-dimensional collaborative constraint model is constructed, and the cost constraints comprise server-free function budget and virtual machine budget; integrating the hyper-heuristic framework into a reinforcement learning algorithm; an improved reinforcement learning algorithm is adopted to solve the cloud workflow scheduling problem, optimal execution resources are selected, and an optimal scheduling scheme is obtained; and performing real-time scheduling according to the optimal scheduling scheme, and introducing a deviation feedback mechanism to monitor an execution error in real time. According to the method, the workflow scheduling problem is decomposed into a closed-loop optimization process of state perception and action decision, dynamic environment perception, multi-target tradeoff and online strategy optimization are deeply fused, the global search function is achieved, local optimization can be achieved, the algorithm complexity is low, and robustness is high.

The invention discloses a production process scheduling method, system, equipment and medium based on the industrial Internet of Things, and relates to the technical field of industrial Internet of Things data processing, and the method comprises the steps: obtaining a to-be-scheduled production process, obtaining a preset production queue which is composed of a current production task and a plurality of to-be-processed production tasks, obtaining completion time and production information; obtaining environment trend information, obtaining a dynamic correction index, obtaining a process utilization rate, and obtaining a process utilization index based on the correction model, the process utilization rate and the dynamic correction index; obtaining inventory unit time cost and delivery deadline, and obtaining a scheduling index based on the scheduling model, the completion time, the process utilization index, the inventory unit time cost and the delivery deadline; and sequentially arranging the plurality of production tasks to be processed according to the scheduling indexes from small to large, generating a scheduling queue, and executing the production tasks according to the scheduling queue. The method has the advantages of dynamic environment self-adaption, multi-target collaboration and good scheduling effect.

The invention relates to the technical field of engineering construction progress intelligent control, in particular to an engineering construction progress intelligent control method and system based on digital twinning, and the method comprises the steps: obtaining a preset plan logic relation set; acquiring a real-time process state set of the construction site; determining a plan failure index through real-time logic deviation calculation; in response to the plan failure index being greater than a preset failure determination threshold, determining that the plan fails and switching to a logic emergence mode; in response to the plan failure index being less than or equal to the failure determination threshold, maintaining the plan driving mode; deducing and generating an emergence logic set through emergence logic confidence calculation; determining a failure logic set; reconstructing to generate a hybrid logic model; the prediction completion time is deduced again; outputting a decision support signal; according to the method, the defect that a traditional system cannot judge the failure is overcome, the prediction accuracy is improved, and invalid warning out of reality is avoided.

The invention provides a deadline constraint-based DAG application dynamic scheduling optimization method and system, and the method comprises the steps: converting an application deadline into an instant reward of each task scheduling through employing a DAG structured encoder, a Transform encoding network based on gating feature fusion, and a multi-action selection deep reinforcement learning task scheduling method based on a pointer network, in combination with a dynamic mask scheme, the mobility of a DAG application and the dynamic nature of edge resources are dealt with, then priority subtask selection and real-time decision of the scheduling position of the priority subtask selection are made, and the completion time and execution energy consumption of the application are reduced. In order to stabilize and accelerate DRL scheduler training, task encoder training and reinforcement learning training are decoupled, and a DAG encoder is pre-trained based on self-supervised learning.

The invention relates to a dual-resource constraint flexible job shop scheduling method for reducing worker load, which comprises the following steps: S1, construction of a worker load model: dividing the worker load model into four conditions of light work, moderate work, micro-severe work and rest according to daily work arrangement of workers, and setting the maximum working time length for each worker, the calculation module is used for calculating extra workloads; in the scheduling method provided by the invention, three different initialization strategies are combined, and the proportion of the initialization strategies in a population is set, so that the diversity and quality of an initial solution are ensured, specifically, the used initialization strategies comprise random initialization, initialization according to the process completion time and initialization according to the process remaining time; under the random initialization strategy, the initial solution of the population has great diversity, which is helpful for avoiding the trouble of a local optimal solution.

The invention discloses a manufacturing task autonomous negotiation and execution method based on a large language model agent, and the method comprises the steps: constructing a production scheduling agent, an equipment management agent, a material distribution agent and a quality control agent, analyzing a natural language task instruction through the production scheduling agent, and decomposing the natural language task instruction into subtasks; each agent calculates a utility value based on the load rate, the resource matching degree, the estimated completion time and the historical success rate, and performs structured negotiation to achieve a task allocation consensus; a prediction-check-rollback architecture is adopted to generate an action instruction sequence, the sequence is compiled into a time Petri network transition sequence, and reachability verification is carried out based on hard security constraints; production environment data is collected in real time to trigger anomaly detection and re-negotiation, and a formalized security verification and causal anti-factual reasoning parameter updating mechanism is introduced. According to the method, unstructured instruction understanding, autonomous task planning, multi-agent collaborative decision and closed-loop optimization are realized, and the problems of real-time performance, safety and interpretability of a large language model in manufacturing control are solved.

The invention relates to the technical field of production management, in particular to a production scheduling optimization system based on data analysis, which comprises a process data trend analysis module, a station resource real-time acquisition module, a scheduling sequence optimization judgment module, a rhythm coordination check module and a scheduling dynamic optimization module. According to the method, the process duration trend is analyzed in real time, continuous growth is accurately recognized, rhythm parameters are dynamically corrected, the stable process rhythm is ensured, efficient resource connection is achieved based on the station task completion time and the matching task and station idle starting time, the task sequence is dynamically adjusted according to the station idle state, and task connection is optimized; the beat proportion of adjacent tasks is verified, the task interval is adjusted, the beat difference is controlled within a reasonable range, beat sudden change is prevented from interfering with production, the rapid response ability to production changes is integrally enhanced, task and resource collaborative optimization is achieved, the overall scheduling efficiency and beat consistency are improved, and the production continuity and the resource utilization rate are improved.

The invention discloses a construction management system and method based on multi-source data analysis, and relates to the technical field of building construction informatization and intelligent management and control, and the method comprises the steps: capturing a front path task completion signal and a subsequent task starting instruction; obtaining preposed task completion time, a dynamic condition type and a threshold parameter; executing time logic verification based on the dynamic condition type, and judging a construction dynamic context ready state; bIM design coordinates, field positioning coordinates and space tolerance parameters of the construction machinery are obtained, and space logic verification is executed by calculating space deviation; when both the time verification and the space verification pass, subsequent task state conversion is allowed, and otherwise, blocking is carried out; and finally generating a structured report containing all verification parameters and results.

The invention relates to the technical field of time sequence data processing and deep learning, in particular to a time sequence filling method and system based on a coarse-to-fine filling normal form. In the invention, a preprocessing module is used for sampling an input sequence to obtain a subsequence rk with the length of Lk, and a regression prediction module is combined with a causal mask to generate a prediction value r'k with the same structure and the same length as the rk based on all subsequences {r1... rk} generated by the preprocessing module; the correction module performs up-sampling on the predicted value r'k to obtain an up-sampling sequence r ''k with the length of T; supplementing missing data of the original sequence based on the r ''k to obtain a corrected sequence X (k + 1); and traversing k = 1... K to obtain a correction sequence X (K + 1) as a final repair completion time sequence. The method overcomes the defects that the time sequence filling mode in the prior art does not consider the unfixed missing rate and missing value block distribution, and is beneficial to improving the filling accuracy.

The invention discloses an artificial intelligence large model development-oriented low-code automatic generation and deployment system, which comprises the following steps of: performing semantic analysis on a natural language modeling demand input by a user, generating a structured modeling target, and constructing and optimizing a task flow chart. The system generates a corresponding low-code modeling instruction according to the optimized task flow chart, completes automatic matching of tasks and resources by adopting an improved shortest completion time scheduling algorithm in combination with the real-time state of heterogeneous computing resources, and automatically deploys the generated modeling tasks to a target computing environment in a containerization manner, so that the modeling efficiency is improved. And unified operation control and process monitoring are carried out, so that automatic closed loop from semantic input to model deployment is realized.

The invention relates to the technical field of engineering project management, in particular to an engineering progress risk early warning method and system based on adaptive threshold learning. Training the preprocessed data by using a C-L algorithm to obtain a classification model and a classification threshold; using the classification model to classify project progress data, constructing a completion time prediction model and a prediction threshold, and predicting and classifying cases lacking completion time; self-adaptively updating the threshold value on the test set through a C-L algorithm until a satisfactory test effect is achieved, and outputting a final prediction threshold value; the self-adaptive threshold mechanism can automatically adjust the early warning threshold according to the characteristics of different engineering projects, and compared with a traditional fixed threshold method, the early warning accuracy is improved by 20-30%.

The invention relates to the technical field of flexible job shop scheduling in intelligent manufacturing and production scheduling, in particular to a flexible job shop scheduling optimization method for solving and adjusting resource constraints. Comprising the steps of initializing parameters and randomly generating an initial population; sequentially using a crossover operator and a mutation operator to evolve the current population; executing a hybrid decoding strategy on the current population; sorting the individuals in the current population from small to large according to the maximum completion time to form an elite population, and updating the elite population through question specific local search; and judging whether an evolution condition is met or not, if so, executing CP-based mathematical evolution, and outputting a final solution when the running time reaches the total running time. The method has the positive effects of reducing the resource waiting time, improving the machine utilization rate and improving the resource utilization efficiency and the scheduling performance of the whole workshop production.

The invention discloses a method for dispatching automatic guided vehicles in a parallel arrangement container terminal. The method comprises the following steps: collecting container terminal AGV dispatching parameters; establishing an AGV scheduling optimization model; solving the AGV scheduling optimization model; and outputting an AGV scheduling scheme. The AGV scheduling optimization model with minimum AGV power consumption and minimum operation completion time as optimization objectives is constructed by considering the influence of AGV path conflicts and the charging process on actual operation, so that collaborative optimization of AGV charging, task assignment and path planning problems is realized, and the method has the characteristics of high universality and wide coverage. According to the method, the improved adaptive genetic algorithm is adopted, and adaptive adjustment of crossover and mutation operators gives consideration to population diversity and convergence rate at the same time. And meanwhile, conflict-free path planning is carried out by adopting a space-time A * algorithm, so that collaborative optimization of AGV task scheduling and path planning is realized, the prediction precision is high, the convergence is fast, and the workload is small.

The invention discloses an underwater acoustic processing CPU-GPU dynamic load balancing method based on a task flow model, and belongs to the technical field of underwater acoustic processing. The method comprises the following steps: deconstructing an underwater acoustic processing application into a task flow model represented by a directed acyclic graph; establishing a feature portrait including calculation complexity, parallelism and data throughput for each task node, and constructing a cost prediction model; the CPU / GPU utilization rate and the data transmission performance are monitored in real time; a processor is distributed to each task node by using a dynamic programming algorithm in combination with a cost prediction model and a real-time system state with the goal of minimizing the total task flow completion time; and dynamically scheduling tasks through a central scheduler according to a decision result, and periodically updating a strategy. According to the method, the defects that static task division lacks adaptability and neglects task dependence and communication overhead are overcome, dynamic and efficient utilization of CPU and GPU resources is achieved, and the efficiency and real-time performance of underwater sound processing are remarkably improved.

The invention discloses a multi-rotating-speed flexible job shop scheduling method applied to double-resource constraint, and the method comprises the steps: taking the maximum completion time and minimum total energy consumption of a minimum machine as target functions, and constructing a flexible job shop scheduling model considering the rotating speed energy consumption of the machine and the production demands of a fine process; a machine speed gear constraint, a fine process constraint, a process sequence constraint, a completion time constraint, a machine processing constraint and a worker operation constraint are established as constraint conditions of the model; the flexible job shop scheduling problem is solved by adopting an improved artificial bee colony algorithm, bee colony search guided by excellent genes is adopted in bee learning operation in the improved artificial bee colony algorithm, and nectar source optimization is carried out based on the searched excellent genes; the following bee operation adopts a neighborhood structure which considers machine speed change and balances the working time of workers to carry out dynamic neighborhood search so as to optimize a nectar source. The effectiveness of the improved strategy is verified through experiments, and the superiority is verified through comparison of different algorithms on expansion standard examples.

According to the motion planning method and device for the multi-mechanical-arm cooperative task, a large number of potential conflicts are actively avoided in the early stage of task allocation through division of a virtual wall and a safety area and a danger area, the difficulty and the calculated amount of follow-up track collaboration are remarkably reduced, and the success rate and the efficiency of planning are improved. Cost and load balance are considered in the task allocation stage; in the path point sorting stage, minimizing the total movement distance and the total completion time is taken as a target; and track generation adopts a time optimal algorithm. The layered decoupling design avoids the huge calculation overhead of centralized planning, and meanwhile, the quality of a final solution is ensured through the optimization strategy of each stage. The innovative cost function takes collision distance into consideration, and guides the planner to select a safer path. Through layered and ordered collision solution strategies, such as deceleration, local re-planning and optimal waiting, it is ensured that a collision-free solution can be always found in a complex dynamic environment, and robustness is high.

The invention provides a packaging test workshop intelligent scheduling algorithm based on multi-scale information fusion driving, and belongs to the field of intelligent optimization scheduling, and the method comprises the steps: firstly analyzing a workshop production process and constraint conditions, and constructing a multi-target optimization model with the target of minimizing the maximum completion time, the total mold change time and the total tardiness time; a scheduling framework fusing an LSTM-Informer prediction model and an ITD3 algorithm is provided, prediction features are generated through multi-source data collection and fusion, multi-scale prediction of a buffer area state is realized by using the LSTM-Informer, and the ITD3 algorithm is driven based on prediction information to carry out dynamic decision making; and meanwhile, a composite scheduling rule and a layered reward mechanism are designed to improve the solving efficiency. According to the method, the dynamic scheduling problem in actual production can be effectively solved, and the maximum completion time, the total die change time and the total tardiness time are remarkably shortened.

The invention relates to the technical field of deep learning, and discloses a fine-grained pipeline scheduling method and device based on cluster node memory difference perception. The method comprises the following steps: estimating the ratio of the peak memory occupancy to the memory capacity of each GPU node, comparing the ratio with a preset threshold value, and selecting a re-calculation strategy and a back propagation segmentation strategy: by taking minimization of the end-to-end training time of a model as a target, scheduling and modeling micro-batch data as a flow shop problem, analyzing the dependency constraint of each flow line stage, and calculating the flow shop problem; comprising forward-back propagation sequence dependence, stage dependence, operation dependence and memory limitation, and generating a micro-batch operation sequence scheduling scheme meeting the memory capacity limitation; according to the scheduling scheme, an execution sequence queue is created by executing sorting, and the execution time sequence of calculation blocks, communication blocks and re-calculation operation is dynamically coordinated. According to the method, the GPU equipment utilization rate can be remarkably improved, and the end-to-end training completion time of the model is shortened.